The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes technical design documents and test cases to identify issues and divergences. This creates an indirect prompt injection surface as these files are ingested into subagent contexts.
Ingestion points: Technical design documents (design.md), test cases (cases.md), and intake files (digest, audit, pm-checklist) in the sdlc/ directory.
Boundary markers: The skill uses structural markers such as ARTIFACT: and CONTRACT: within its adversarial prompt templates to delimit external data from instructions.
Capability inventory: The skill executes local Python validation scripts and utilizes MCP tools for database and code graph exploration. It has write access to the sdlc/ directory for generating review reports.
Sanitization: No explicit string sanitization is performed on the ingested text, but the skill relies on structured template matching and regex-based validation scripts (check_trace.py) to maintain integrity.
[COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/check_trace.py and scripts/guard_dev.py) to perform mechanical verification of review outcomes and gate-keeping logic.
The scripts are part of the skill's local distribution and use standard libraries (re, pathlib, sys) for path resolution and markdown parsing.
Execution is limited to the local project workspace and follows the intended primary purpose of the skill.

sdlc-gate